[R-sig-ME] glmmTMB dispformula

[Sainsbury, Nigel]

Dear List,

I would greatly appreciate any guidance the community can provide relating to the way the dispformula argument works in glmmTMB.

Context:

I am running a Gaussian mixed model in glmmTMB to account for heteroscedasticity of residuals caused by two fixed effects (one continuous, one categorical). My model code therefore includes the argument, dispformula = ~ X1 + X2, where X1 and X2 are the two fixed effects in question (as per Brooks at al., 2007).

Problem:


1.      Despite searching glmmTMB package documentation and articles, email lists and Stack Overflow, I have been unable to discover how the dispformula argument manages heteroscedasticity in the model i.e. how it connects the covariates (X1 and X2 in this case) to the residual variance. For fixed effect only linear models, Zuur et al. (2009) Chapter 4 provides an excellent explanation of how the gls function in nlme can use different variance structures to control for heteroscedasticity (e.g.  VarFixed (Fixed variance), VarIdent (Different variances per stratum), VarPower (Power of the variance covariate) etc.). Can anyone explain this for me?



2.      If the specification of dispformula does explain the heteroscedasticity in the fixed effects, should the plots of model residuals against fitted values and X1 and X2 appear homoscedastic or remain heteroscedastic as when the model is run using lme4 i.e. the dispformula argument accounts for, but does not remove, the heteroscedasticity?


Thanks in advance for any guidance on this issue.

Best wishes,
Nigel

	[[alternative HTML version deleted]]